Oil Discoveries and Political Windfalls: Evidence on Presidential Support in Uganda Laura Paler * Jeremy Springman † Guy Grossman ‡ Jan Pierskalla § June 22, 2020 Abstract It is widely believed that oil discoveries cause bad governance and conflict. However, research on the political resource curse argues that oil often increases support for incumbent chief executives while the conflict curse literature suggests it erodes it, especially when discovered in opposition areas. We draw on research on distributive politics to theorize how the effects of oil on incumbent support will vary depending on whether it is discovered in core, swing, or opposition constituencies. Our findings, based on electoral and survey data from Uganda and a difference-in-differences design with heterogeneous effects, show that differential voter responsiveness to targeted oil benefits increased support for the incumbent when oil is discovered in swing constituencies. Ultimately, we highlight how the local political context shapes the effect of oil on the strength of support for the incumbent chief executive, with important implications for understanding the roots of both the political and conflict curses. * University of Pittsburgh (email: [email protected]). † Duke University & DevLab@Duke (email: [email protected]). ‡ University of Pennsylvania (email: [email protected]). § Ohio State University (email: [email protected]). We would like to thank Leslie Marshall for her excellent research assistance and Lucy Martin, Michael Ross, and William Spaniel for their valuable com- ments. The original nationally representative survey was approved by IRB at the University of Pennsylvania (Protocol #818702).
42
Embed
Oil Discoveries and Political Windfalls: Evidence on ...
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Oil Discoveries and Political Windfalls:Evidence on Presidential Support in Uganda
Laura Paler∗ Jeremy Springman† Guy Grossman‡ Jan Pierskalla§
June 22, 2020
Abstract
It is widely believed that oil discoveries cause bad governance and conflict. However,research on the political resource curse argues that oil often increases support forincumbent chief executives while the conflict curse literature suggests it erodes it,especially when discovered in opposition areas. We draw on research on distributivepolitics to theorize how the effects of oil on incumbent support will vary depending onwhether it is discovered in core, swing, or opposition constituencies. Our findings, basedon electoral and survey data from Uganda and a difference-in-differences design withheterogeneous effects, show that differential voter responsiveness to targeted oil benefitsincreased support for the incumbent when oil is discovered in swing constituencies.Ultimately, we highlight how the local political context shapes the effect of oil on thestrength of support for the incumbent chief executive, with important implications forunderstanding the roots of both the political and conflict curses.
excellent research assistance and Lucy Martin, Michael Ross, and William Spaniel for their valuable com-ments. The original nationally representative survey was approved by IRB at the University of Pennsylvania(Protocol #818702).
1 Introduction
Recently, a number of oil and gas explorations culminated in dramatic discoveries in several
East African countries. In Tanzania, deep offshore explorations have led to the discovery of
at least 45 trillion cubic feet of natural gas;1 in Kenya, a 2012 oil discovery in the Turkana
region resulted in the country’s first crude exports in 2019;2 Rwanda and Mozambique have
recently discovered commercial quantities of natural gas and are actively exploring for oil
reserves;3 and in 2019, significant new discoveries were made in South Sudan.4
Discoveries of oil and natural gas raise important questions about how new resource en-
dowments impact political and economic development, especially in low-income countries.
While resource wealth has the potential to benefit these countries, all too often it leads to
undesirable outcomes. A vast literature on the ‘political’ resource curse shows that resource
windfalls often strengthen autocratic rule (Beblawi and Luciani, 1987; Ross, 2001), weak-
ens accountability in democracies (Aslaksen, 2010; Ramsay, 2011; Paler, 2013; Brollo et al.,
2013), and increases corruption and patronage (Robinson, Torvik and Verdier, 2006). Re-
search on the ‘conflict’ curse emphasizes how natural resource wealth increases the likelihood
or duration of separatist or center-seeking civil war by igniting societal grievances, providing
the material incentives or means to engage in violent rebellion, and inducing the emergence
of political challengers (Humphreys, 2005).5
Importantly, past work on the political and conflict resource curses suggest that oil has
1“Oil and gas discoveries place Tanzania within the crosshairs of parasitic imperialism,” Uhuru
News, April 8, 2014.
2“Kenya’s first crude oil export sparks demands over revenue sharing,” Reuters, August 26,
2019.
3“Rwanda pauses oil exploration in Kivu,” The East African, October 22, 2018.; “Rwanda
pauses oil exploration in Kivu,” Reuters, October 8, 2018.
4“South Sudan Discovers New 300 Million Barrels of Oil,” Chimp Reports, August 21, 2019.
5For a review of both the political and conflict curse literatures, see Ross (2013).
where β3, β10, and β11 are the main coefficients of interest and jointly capture the average
marginal effect of the oil discoveries as well as the effects of the oil discoveries within core,
swing, and opposition localities. We note that while we aim to identify a causal effect of the
oil discoveries, the conditioning variable voteshare is not itself causally identified. While the
inclusion of fixed effects in our main specification controls for potential confounders, we also
conduct robustness checks (described below) in which we use a matching strategy to address
concerns about possible confounding.
Causal inference in our estimation approach derives from the assumption that trends
in incumbent support in oil and non-oil core, swing, and opposition localities would have
been the same in the absence of the oil discovery. In other words, within constituencies
designated as core, swing, or opposition, there would have been no divergence in incum-
bent support trends across oil versus non-oil constituencies had oil not been discovered. We
present evidence to support the parallel trends assumption across oil and non-oil localities
in general as well as within core, swing, and opposition localities using geo-tagged responses
from Afrobarometer including three pre-treatment (2000, 2002, 2005) rounds. We use Afro-
barometer data rather than election results since the presidential elections in 2006 were the
first in which opposition parties were allowed to compete. Figure 2 shows average approval of
President Museveni’s job performance for respondents in oil and non-oil villages overall and
disaggregated into core, swing, and opposition villages.21 In each case, trends in support for
21The question asks: “Do you approve or disapprove of the way that the following people have
20
0.25
0.50
0.75
1.00
Vote Share
0.25
0.50
0.75
1.00
Vote Share
Unpopulated
Missing
Restricted
Figure 1: This figure shows pre-treatment variation in President Museveni’s vote share inparishes that are within 100km (red scale) of an oil discovery and parishes that are not (bluescale). Dark grey parishes indicate parishes that are more than 200km from an oil discovery.
Museveni were similar prior to the discovery in 2006. The figures also indicate a divergence
in the post-2006 period, which is what we estimate formally.
Unbiased causal inference in our setup requires that a number of other assumptions hold,
including that our treatment is not confounded with other possible ‘treatments’ underway at
the same time and that there are no omitted time-varying variables that could differentially
drive changes in incumbent support in core, swing, or opposition localities. We investigate
these below as robustness checks to our main results.
performed their jobs over the past twelve months, or havent you heard enough about them to
Figure 2: Panels show trends in average approval of President Museveni’s job performance(4-point scale) over time across core, opposition, and swing areas. ‘Oil’ respondents arelocated in villages within 100km of the nearest oil discovery and control respondents are invillages 100–200km from the nearest discovery. Dashed vertical lines indicate the month ofthe 2006 and 2011 general elections. The solid vertical line indicates the month of the firstreported oil discovery. 2006 vote share is measured according to the vote share or the parishin which each village is located. Source: Afrobaromter rounds 1–6.
22
5 Main Results
We first examine the average marginal effect of the oil discovery on presidential support.
Figure 3 shows the average effect of proximity to an oil discovery on Museveni’s win margin
and vote share, with corresponding regression results in Appendix C. As discussed in Section
2, it is difficult to predict whether oil discoveries will have a positive, negative, or null effect
on incumbent support on average in light of the heterogeneity by local political context
theorized above. Nevertheless, we find that oil discoveries increased President Museveni’s
vote share by about 1.6 percentage points on average, an effect that is statistically significant
at the 95 percent confidence level. The effect size is also substantively meaningful, showing,
for instance, an increase in Museveni’s win margin by over four percentage points. Overall,
Museveni has gained in oil, relative to non-oil, constituencies after 2006.
●
●
0.00
0.02
0.04
0.06
Win Margin Vote Share
AM
E (
Tre
atm
ent
< 1
01km
)
Average Marginal Effect of Oil Discovery on Electoral Support for the President (2006 − 2011)
Figure 3: Difference-in-differences estimates for the average marginal effect of the oil discov-ery on electoral support for President Museveni for the 2006 and 2011 elections. Treatedparishes are located within 100km of the nearest oil discovery and control parishes are within101–200km.
Moreover, as expected, this average effect obscures significant variation in the effect of
oil discoveries when disaggregating results by constituency type. Figure 4 presents our main
results, showing the pre-post effect of the oil discovery across the range of the continuous
23
measure of 2006 vote share. We find that the biggest pre-post increases in win margin and
vote share for Museveni across oil versus non-oil parishes occurred in swing parishes, or those
parishes in the range from 40–60 percent vote share for the NRM in the 2006 election. This
political windfall is also substantively large. At its peak, the gains for the president represent
an almost 10 percentage point increase in vote share in the wake of the oil discovery. These
findings for swing areas contrast with those for core and opposition areas. For core areas
(above 60 percent vote share), we see modest positive gains that are decreasing as vote share
increases (consistent with a ceiling effect). In opposition areas, the effects of oil become
slightly positive in the 20-40 percent range, consistent with the notion that there is room
for oil benefits to persuade opposition loyalists to switch support or to increase support
among core or swing voters within opposition constituencies. The inverted U-shape of the
results suggest that the effects of oil discoveries in swing areas are indeed different from
those in core or opposition localities. Results using a binned estimator recommended by
Hainmueller, Mummolo and Xu (2019) and reported in Appendix C.2, reinforce that the
effects of oil discoveries in swing areas are different from those in core constituencies.
Strikingly, we find similar results when using the Afrobarometer survey data. We employ
the measure of job approval used to assess parallel trends in Section 4 from the 2005 and 2008
waves (the waves conducted immediately before and after the discovery).22 Figure 5 shows
a very similar pattern to the one above: the oil discovery has a bigger effect on increasing
incumbent support in swing relative to core or opposition localities. While the curve peaks
around 35 percent vote-share, which is borderline opposition/swing, the inverted U-shape is
unmistakable.23
22See Appendix D for a detailed description of the estimation and evidence that we obtain the
same results when including multiple pre and post-treatment waves, alternative bandwidths,
and a battery of covariates.
23The Afrobarometer data is a repeated cross-section rather than panel, and therefore does
not provide the same causal leverage as the election data results.
24
−0.4
−0.2
0.0
0.2
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
Win
Mar
gin
−0.4
−0.2
0.0
0.2
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
Vot
e Sh
are
Marginal Effect of Oil Discovery on Electoral Support for the President (2006−2011)
2006 Vote Share
−0.4
−0.2
0.0
0.2
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
Win
Mar
gin
−0.4
−0.2
0.0
0.2
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
Vot
e Sh
are
Marginal Effect of Oil Discovery on Electoral Support for the President (2006−2011)
2006 Vote Share
Figure 4: Triple Difference-in-difference estimates for the marginal effect of the oil discoveryon electoral support for President Museveni across levels of pre-treatment support. Treatedparishes are located within 100km of the nearest oil discovery and control parishes are within101–200km, and results are presented for the 2006 and 2011 elections. Shaded region is the95% confidence interval. Data rug indicates the distribution of treatment parishes by theirpresidential vote share in 2006.
25
−1
0
1
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
Marginal Effect of Oil Discovery on Presidential Job Approval
2006 Vote Share
Ave
rage
Mar
gina
l Eff
ect
−1
0
1
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
Marginal Effect of Oil Discovery on Presidential Job Approval
2006 Vote Share
Ave
rage
Mar
gina
l Eff
ect
Figure 5: Marginal effect of the oil discovery on Museveni’s job approval across levels ofpre-treatment support. Treated parishes are within 100km of an oil discovery. The sample isrestricted to exclude responses from parishes more than 200km from the oil. Shaded regionis the 95% confidence interval. Data rug plots indicate the distribution of treatment parishesby their presidential vote share in 2006.
5.1 Robustness Checks
We increase confidence in the results presented above by conducting a number of robustness
checks. We first test the robustness of results to different ways of defining the treatment and
control groups. We test the sensitivity of our bandwidth to five different distances between
75km and 115km from the nearest oil well (with a corresponding adjustment to the control
group to remain two times the distance of the treatment group). The results in Appendix
E.1 show that our findings are highly robust to the precise cutoff.
It could also be that the observed changes preceded the oil discovery in 2006 such that
we are not detecting the effects of the oil discoveries but rather changes that were underway
for different reasons. We examine this by implementing a simple placebo test using the
Afrobarometer measure of job approval in which we restrict our analysis to pre-oil waves
(2000, 2002, and 2005) and produce a placebo treatment assignment indicator that defines
treatment as occurring between 2002 and 2005. As shown in Appendix E.2, this placebo
Bias in our estimates could also arise if there were time variant factors driving differential
changes in incumbent support in oil and non-oil localities within core, swing, or opposition
areas. While our support for the parallel trends assumption helps to alleviate these concerns,
it cannot completely eliminate them. To address this, we use a rich range of parish-level
covariates drawn from the 2002 census and implement an entropy weighting matching algo-
rithm to improve balance in covariates across oil-and non-oil constituencies, helping to ensure
that the control group is an appropriate counterfactual to the treatment group (Hainmueller,
2012). As reported in Appendix C, we obtain the same patterns when using the entropy
weighted data.
Another assumption for causal inference is that there was no other shock during our
treatment period that could have differentially affected the western region of the country
where our oil localities are predominantly located. One potential concern is that Uganda’s
oil is located in the region that was also most affected by the insurgency propagated by
the Lord’s Resistance Army, which ended with a truce in August 2006. It is possible that
the changes we are detecting are due to developments in the immediate post-conflict period
rather than oil. This seems unlikely insofar as many of our control localities in the north
were also affected by the conflict and therefore would also have been affected by time-varying
factors related to the end of hostilities. Nevertheless, we conduct two additional tests, ruling
out the possibility that our results are driven by more aid rather than by the oil discoveries
(see Appendix E.3) or by differential migration into (or out of) swing oil localities following
the end of the conflict (see Appendix E.4).
We also conduct a number of additional tests to ensure that our outcome measure of
electoral support for Museveni is capturing the full range of dynamics theorized in Section
2. One possibility is that the results are being driven by changes in support for the NRM
rather than by changes in support for the president himself. Drawing on Afrobarometer data
on self-reported identification with the NRM, we find only weak evidence that oil discoveries
27
increased identification with the NRM in swing communities (see Appendix E.5). This
reinforces that gains went primarily to the chief executive.
We also consider whether the weaker results in core and opposition areas might be because
changes in incumbent support are manifesting in ways other than changes in vote share. For
instance, as suggested by theories of distributive politics, it could be that the primary effect
of strategic promises to core areas is increased turnout. We show in Appendix E.6 that
there is little evidence that oil discoveries affected turnout in core, swing, or opposition
constituencies. For opposition areas, it is possible that the oil discovery affected incumbent
support not as expressed through electoral politics but rather through unconventional modes
of participation like protests and violence. If so, we might be missing changes in incumbent
support not expressed at the ballot box. We examine in Appendix E.7 whether the oil
discoveries differentially affected protest activity and violence in core, swing, or opposition
localities and find little evidence of this.
All in all, the results presented above provide robust evidence that oil differentially in-
creased incumbent support in swing relative to core or opposition oil constituencies. In what
follows, we provide evidence for our preferred mechanism—that these changes in incumbent
support reflect voter responsiveness to strategic promises of future oil benefits.
6 Support for the Distributive Politics Mechanism
Our theory suggests that differential increases in presidential support in swing relative to core
or opposition localities should be driven by greater voter responsiveness to promised benefits
under either symmetric or asymmetric targeting. Given our focus on the discovery period,
it is not possible to examine actual transfers. We thus triangulate data from a number of
sources.
First, we examine whether the oil discoveries induced Museveni to make more promises
to the oil region. To do this, we collected newspaper data on district-specific campaign
28
promises and visits by Museveni made in 2005 (prior to the 2006 elections) and in 2010
(prior to the 2011 elections).24 We find that between the 2005 and 2011 election campaigns,
the president increased promises to non-oil districts (those between 100–200km from an
oil discovery) by 155 percent while increasing promises to districts within 100km of an oil
discovery by 321 percent. Similarly, the president increased visits to non-oil districts by
15 percent while increasing visits to oil districts by 44 percent (see Appendix F.1 for more
detail). This lends support to the notion that the oil discoveries induced the kinds of strategic
promises theorized in Section 2. One limitation of the newspaper data is that we do not
have enough observations to look at differential promises to oil versus non-oil swing, core,
and opposition localities.25 We thus look for further evidence by focusing on expectations
at the individual-level.
Expectations are important to examine because higher expectations could drive incum-
bent promises; alternatively they could reflect the belief that promises made are credible.
Regardless, if our distributive politics logic is correct, we would expect to see higher ex-
pectations of future benefits in oil versus non-oil swing constituencies as well as relative to
core or opposition oil localities. We examine this using original survey data collected from
a nationally representative in-person survey of 2,714 Ugandans conducted in 2014 (see Ap-
pendix F.2 for details on the survey methodology). While this data is observational and
cross-sectional and as such does not capture the causal effects of the oil discovery, it allows
us to examine descriptively how expectations differ across oil and non-oil core, swing and
opposition localities.
We focus on questions from the survey that speaks most immediately to respondents’
24Due to the nature of media coverage of presidential campaigning, it was infeasible to collect
these data at a lower administrative level than the district. Because there were only 36
districts with centroids within 200km of an oil discovery, we rely on descriptive tables rather
than regression models.
25There were only four swing districts within 200km of an oil discovery.
29
N = 2630N = 2630N = 2630N = 2630
●
●
N = 376N = 376N = 376N = 376
●
●
N = 770N = 770N = 770N = 770
●
●
N = 1484N = 1484N = 1484N = 1484
●
●
Swing Core
All Respondents Opposition
0.2 0.4 0.6 0.2 0.4 0.6
You or YourHousehold
Uganda
You or YourHousehold
Uganda
Share Respondents Who Answered "Very Big"
● OilControl
Question: Do you think that the benefits from oil for [...] will be [...]?Respondent Views on the Benefit Of Oil
Figure 6: Average responses from a nationally representative survey of 2,714 people from 76constituencies and 304 villages in 52 districts across Uganda.
views about whether they will benefit from oil. Specifically, the survey included two questions
that asked: “Do you think that the benefits from oil for [Uganda/for you or your household]
will be very big, somewhat big, not too big, not big at all?” The results presented in Figure
6 show that individuals in swing oil constituencies are more likely than those in non-oil
swing constituencies to say that they expect ‘very big’ benefits. Moreover, the difference
in expectations between oil versus non-oil swing constituencies is bigger than that in oil
versus non-oil core or opposition localities.26 These findings suggest that voters in swing oil
constituencies expected to receive more benefits and, consequently, might have been more
willing to reward the incumbent than those in swing localities without oil.
To gain more leverage on whether the oil discovery caused bigger increases in incumbent
support in swing relative to core or opposition areas, we turn to Afrobarometer data. Specif-
ically, we implement our difference-in-differences estimation with heterogeneous effects with
dependent variables that capture satisfaction with government performance on the delivery
26We also find a stronger sense of ownership overall in oil localities, see Appendix F.2.
30
of public goods, namely education, health care, sanitation, and food security.27 The results
in Figure 7 shows the the same inverted U-shape that we find in our main results. Specifi-
cally, we find that the oil discoveries caused bigger increases in satisfaction with the delivery
of all four goods in swing relative to core or opposition areas. We obtain a similar pattern
when using measures of improved perceptions of household welfare (see Appendix F.3).
−2
−1
0
1
2
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
Index
−1.0
−0.5
0.0
0.5
1.0
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
Education
−1.0
−0.5
0.0
0.5
1.0
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
Health
−2
−1
0
1
2
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
Food
−2
−1
0
1
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
Sanitation
Marginal Effect of Oil Discovery on Public Goods Satisfaction
2006 Vote Share
Ave
rage
Mar
gina
l Eff
ect
−2
−1
0
1
2
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
Index
−1.0
−0.5
0.0
0.5
1.0
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
Education
−1.0
−0.5
0.0
0.5
1.0
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
Health
−2
−1
0
1
2
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
Food
−2
−1
0
1
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
Sanitation
Marginal Effect of Oil Discovery on Public Goods Satisfaction
2006 Vote Share
Ave
rage
Mar
gina
l Eff
ect
Figure 7: Marginal effect of the oil discovery on perceptions of government performance inpublic goods provision across levels of pre-treatment support. Treated parishes are locatedwithin 100km of the nearest oil discovery and control parishes are within 101–200km. Shadedregion is the 95% confidence interval.
27Because Afrobarometer is a repeated cross-section (rather than panel), we drop unit fixed
effects. The relevant questions asks “How well or badly would you say the current government
is handling the following matters: Improving basic health services, Addressing educational
needs, Providing water and sanitation services, Ensuring everyone has enough to eat.”
31
However, using a wide variety of data sources, we do not find evidence that the oil dis-
coveries resulted in increases in actual goods provision or benefits for swing areas in the
discovery period. Additional analysis in Appendix F.4 shows that the results above are not
mirrored in Afrobarometer data on actual access to public services at the village and sub-
county levels. We also find no evidence of increased access to health care or improved health
outcomes using household, mother, and birth-level data from Uganda’s 2006 and 2011 De-
mographic Health Surveys. Similarly, neither 2008 and 2013 data from the Ugandan Bureau
of Statistics on roads and highways, nor 2006 and 2013 nightlights data from NOAA, reveal
differential improvements in infrastructure or economic development in swing constituencies
proximate to oil discoveries. We also find no evidence swing oil constituencies were more
likely to receive district status than their non-oil counterparts (see Appendix F.4). This null
result is telling because splitting larger administrative units into smaller ones is a tactic com-
monly used in Uganda and elsewhere to credibly commit to future transfers to new political
groups from which incumbents are trying to win political support (Gottlieb et al., 2019).
All in all, our results indicate that oil discoveries increased presidential promises to oil
localities; that swing oil localities harbor the highest expectations of future oil benefits;
and that satisfaction with public goods provision increased more in swing relative to core
or opposition oil localities, even in the absence of changes in actual public goods provision.
While it might seem surprising that we observe increases in expectations and satisfaction
even without commensurate increases in actual benefits, these results are consistent with the
logic of distributive politics laid out in Section 2. Our theory predicts that, in the discovery
period, voters in swing oil areas will differentially reward the incumbent for credible promises
of future oil benefits. It is also possible that respondents in swing oil areas are over-estimating
improvements in public goods provision as a result of stronger identification with the chief
executive (Carlson, 2016).
We also consider alternative explanations for our key results. One possibility is that
voters in swing oil constituencies engaged in ‘strategic signalling’—preemptively voting for
32
Museveni in 2011 in order to send a credible signal of willingness to reward for future oil
benefits. Strategic signalling, often referred to as ‘voting wisely’, is relatively common in
Uganda.28 Such behavior might be particularly effective at attracting promises from Musev-
eni when it occurs in swing constituencies. While strategic signaling could explain why we
find changes in incumbent support before any increases in actual benefits, we emphasize that
this mechanism is consistent with our distributive politics story. While strategic signalling
alters which player moves first, voters and incumbents are still playing a distributive politics
game and the same predictions apply: oil discoveries will result in differential increases in
incumbent support to swing relative to core or opposition localities due to voters’ unique
responsiveness to expected future oil benefits.
It is also possible that there are other, non-distributive politics mechanisms that could
explain our results. It could be that bigger increases in incumbent support in oil-rich swing
localities were driven not by promises of future benefits but rather by more threats and
intimidation. Yet, intimidation is not consistent with the newspaper evidence that Museveni
made more promises to oil constituencies. We also find no evidence (albeit self-reported)
that the oil discovery increased concerns about electoral fairness or reduced trust in the
electoral commission in swing oil constituencies (see Appendix F.6). All in all, while we
cannot entirely rule out intimidation as an alternative mechanism, it does not necessarily
better explain the observed pattern of results than the distributive politics story.
7 Conclusion
Research on the resource curse remains centrally concerned with understanding why oil
strengthens incumbent governments in some contexts while weakening them through unrest
and violent rebellion in others. The main contribution of this paper is to show how the
political context at the local level—namely the extent to which constituencies have histori-
28“Museveni to Masaka: vote wisely to get roads,” The Observer, December 18, 2018.